Modelling recorded crime: a full search for cointegrated models

TL;DR

This paper introduces a model generator that efficiently searches for cointegrated models, demonstrating its application in modeling recorded violent crime in the Netherlands and showing its superior forecasting performance over short-run models.

Contribution

A novel model generator that automates the search for cointegrated models using the Engle-Granger procedure and information criteria.

Findings

Cointegrated relation between violent crime, demographic, and economic variables.

The generator successfully identifies models with better in-sample forecasts.

The cointegrated model outperforms short-run models in forecasting accuracy.

Abstract

A modelgenerator is developed that searches for cointegrated models among a potentially large group of candidate models. The generator employs the first step of the Engle-Granger procedure and orders cointegrated models according to the information criterions AIC and BIC. Assisted by the generator, a cointegrated relation is established between recorded violent crime in the Netherlands, the number of males aged 15-25 years (split into Western and non-Western background) and deflated consumption. In-sample forecasts reveal that the cointegrated model outperforms the best short-run models.